High-quality Application Layer Multicast System Based On Hybrid-mesh Overlay

Applying streaming technology into high-quality video service for large-scale Internet users becomes one of hot spots between computer network and Multimedia. In the framework (model) of Application Layer Multicast (ALM), Overlay Network, P2P, Media Coding, Network Coding and other theories or technologies play an important role in promoting large-scale streaming application.Whether adopts structured approach to build overlay network is the first question of ALM, and thus extends two different overlay models:structured and unstructured overlay network model. Structured model focuses on the building and optimization of topology, while the unstructured model emphasizes scheduling strategy. The former has lower latency and control overhead than the latter, but performance gap between them in both robustness and scalability is significant. Therefore, from a point of view of performance complementarity, integration of the two may bring "1+1> 2" potential performance increse.For bridging the possibility and the actuality, we propose ALM model based on Hybrid-mesh Overlay (HMO-ALM). Structured-mesh overlay (SMO) serves as the backbone of this model, and unstructured mesh overlay (UMO) as the auxiliary. The dissertation covers the key technologies of HMO-ALM. It is aimed at addressing high-quality (integration of efficiency, robustness and scalability) issues and improving Qos of streaming. The major contributions are as follows.(1) SMO based ALM model is proposed to build backbone. The model forms the hierachical clustering structure according to the definition of self-similar structured graph. Combined with the initial data flow diagram, it forms balanced structure similar to multi-trees. It has a better efficiency and load-balance than traditional hierarchical and multi-tree model, and the impact of node failure and control costs are controlled within a narrow range. In particular, structured overlay optimization is based on cluster replacement as a whole, and the ratio of benefits to overhead cost is more than single node optimization methods. To deal with node failures, the corresponding mechanisms of linear network coding, active link repairing and cluster splicing is designed. It maintains a stable performance in a variety of failure by the comprehensive use of three kinds of mechanism.(2) Collaborative construction mechanisms of UMO and cross-layer scheduling strategy is propsed in auxiliary. QoS aware clustering methods are used to bulid and optimize UMO by collaborating with backbone. Comparing with other QoS aware and random neighbor methods, optimizing speed is improved obviously, and average link delay is smaller. Stability, bandwidth contribution and network proximity as three key factors of node performance evaluation is proposed to support node cross-layer scheduling. Based on the optimal performance configuration parameters of three factors, the strategy significantly improves effective download rate. To schedule cross-layer stream seamlessly, substream-driven and sub-optimal scheduling model and distributed algorithm are proposed. The algorithm has a great advantage over data-driven and hybrid push-pull scheduling algorithm in costs (control costs and costs of content redundancy), delay and throughput.(3) A live streaming system. based on HMO-ALM, ClusterStream, is implemented and staged scheduling mechanism is proposed to meet the requirement of interactive application, which requires both delay-guaranteed and good scalability. Staged scheduling mechanism provides delay guarantees, as well as effectively reduces the costs of content redundancy. Prototype experiments in campus environment show that ClusterStream has a great advantage in the metrics of delay (startup delay and playback delay), broadcast quality and costs, consistent with experimental results.(4) A Hybrid-Mesh overlay solution is proposed to address issues of large-scale VoD service. This solution includes cache lookup mechanism based on regular structured mesh and heuristic prefetching policy combining with substream and scalable encoding. The results show that cache lookup mechanism can improve search efficiency and reduce server stress; prefetching strategies within the larger-scale request, suffers unsharp decline in the quality of video.In a word, this dissertation puts hybrid-mesh overlay network based application layer multicast as research objects, and focuses on studying overlay construction and cross-layer scheduling problems from a point of view of model and system. A series of effective solutions are proposed to address above problems, and a large number of simulation and prototype experiments show the effectiveness of these solutions.